Metric Learning to Accelerate Convergence of Operator Splitting Methods for Differentiable Parametric Programming

Recent work has shown a variety of ways in which machine learning can be used to accelerate the solution of constrained optimization problems. Increasing demand for real-time decision-making capabilities in applications such as artificial intelligence and optimal control has led to a variety of approaches, based on distinct strategies. This work proposes a novel approach to learning optimization, in which the underlying metric space of a proximal operator splitting algorithm is learned so as to maximize its convergence rate. While prior works in optimization theory have derived optimal metrics for limited classes of problems, the results do not extend to many practical problem forms including general Quadratic Programming (QP). This paper shows how differentiable optimization can enable the end-to-end learning of proximal metrics, enhancing the convergence of proximal algorithms for QP problems beyond what is possible based on known theory. Additionally, the results illustrate a strong connection between the learned proximal metrics and active constraints at the optima, leading to an interpretation in which the learning of proximal metrics can be viewed as a form of active set learning.

Learning Constrained Optimization with Deep Augmented Lagrangian Methods

Learning to Optimize (LtO) is a problem setting in which a machine learning (ML) model is trained to emulate a constrained optimization solver. Learning to produce optimal and feasible solutions subject to complex constraints is a difficult task, but is often made possible by restricting the input space to a limited distribution of related problems. Most LtO methods focus on directly learning solutions to the primal problem, and applying correction schemes or loss function penalties to encourage feasibility. This paper proposes an alternative approach, in which the ML model is trained instead to predict dual solution estimates directly, from which primal estimates are constructed to form dual-feasible solution pairs. This enables an end-to-end training scheme is which the dual objective is maximized as a loss function, and solution estimates iterate toward primal feasibility, emulating a Dual Ascent method. First it is shown that the poor convergence properties of classical Dual Ascent are reflected in poor convergence of the proposed training scheme. Then, by incorporating techniques from practical Augmented Lagrangian methods, we show how the training scheme can be improved to learn highly accurate constrained optimization solvers, for both convex and nonconvex problems.

End-to-End Learning for Fair Multiobjective Optimization Under Uncertainty

Many decision processes in artificial intelligence and operations research are modeled by parametric optimization problems whose defining parameters are unknown and must be inferred from observable data. The Predict-Then-Optimize (PtO) paradigm in machine learning aims to maximize downstream decision quality by training the parametric inference model end-to-end with the subsequent constrained optimization. This requires backpropagation through the optimization problem using approximation techniques specific to the problem's form, especially for nondifferentiable linear and mixed-integer programs. This paper extends the PtO methodology to optimization problems with nondifferentiable Ordered Weighted Averaging (OWA) objectives, known for their ability to ensure properties of fairness and robustness in decision models. Through a collection of training techniques and proposed application settings, it shows how optimization of OWA functions can be effectively integrated with parametric prediction for fair and robust optimization under uncertainty.

Learning Fair Ranking Policies via Differentiable Optimization of Ordered Weighted Averages

Learning to Rank (LTR) is one of the most widely used machine learning applications. It is a key component in platforms with profound societal impacts, including job search, healthcare information retrieval, and social media content feeds. Conventional LTR models have been shown to produce biases results, stimulating a discourse on how to address the disparities introduced by ranking systems that solely prioritize user relevance. However, while several models of fair learning to rank have been proposed, they suffer from deficiencies either in accuracy or efficiency, thus limiting their applicability to real-world ranking platforms. This paper shows how efficiently-solvable fair ranking models, based on the optimization of Ordered Weighted Average (OWA) functions, can be integrated into the training loop of an LTR model to achieve favorable balances between fairness, user utility, and runtime efficiency. In particular, this paper is the first to show how to backpropagate through constrained optimizations of OWA objectives, enabling their use in integrated prediction and decision models.

Disparate Impact on Group Accuracy of Linearization for Private Inference

Ensuring privacy-preserving inference on cryptographically secure data is a well-known computational challenge. To alleviate the bottleneck of costly cryptographic computations in non-linear activations, recent methods have suggested linearizing a targeted portion of these activations in neural networks. This technique results in significantly reduced runtimes with often negligible impacts on accuracy. In this paper, we demonstrate that such computational benefits may lead to increased fairness costs. Specifically, we find that reducing the number of ReLU activations disproportionately decreases the accuracy for minority groups compared to majority groups. To explain these observations, we provide a mathematical interpretation under restricted assumptions about the nature of the decision boundary, while also showing the prevalence of this problem across widely used datasets and architectures. Finally, we show how a simple procedure altering the fine-tuning step for linearized models can serve as an effective mitigation strategy.

Projected Generative Diffusion Models for Constraint Satisfaction

Generative diffusion models excel at robustly synthesizing coherent content from raw noise through a sequential process. However, their direct application in scenarios requiring outputs to adhere to specific, stringent criteria faces several severe challenges. This paper aims at overcome these challenges and introduces Projected Generative Diffusion Models (PGDM), an approach that recast traditional diffusion models sampling into a constrained-optimization problem. This enables the application of an iterative projections method to ensure that generated data faithfully adheres to specified constraints or physical principles. This paper provides theoretical support for the ability of PGDM to synthesize outputs from a feasible subdistribution under a restricted class of constraints while also providing large empirical evidence in the case of complex non-convex constraints and ordinary differential equations. These capabilities are demonstrated by physics-informed motion in video generation, trajectory optimization in path planning, and morphometric properties adherence in material science.

Analyzing and Enhancing the Backward-Pass Convergence of Unrolled Optimization

The integration of constrained optimization models as components in deep networks has led to promising advances on many specialized learning tasks. A central challenge in this setting is backpropagation through the solution of an optimization problem, which often lacks a closed form. One typical strategy is algorithm unrolling, which relies on automatic differentiation through the entire chain of operations executed by an iterative optimization solver. This paper provides theoretical insights into the backward pass of unrolled optimization, showing that it is asymptotically equivalent to the solution of a linear system by a particular iterative method. Several practical pitfalls of unrolling are demonstrated in light of these insights, and a system called Folded Optimization is proposed to construct more efficient backpropagation rules from unrolled solver implementations. Experiments over various end-to-end optimization and learning tasks demonstrate the advantages of this system both computationally, and in terms of flexibility over various optimization problem forms.

On The Fairness Impacts of Hardware Selection in Machine Learning

In the machine learning ecosystem, hardware selection is often regarded as a mere utility, overshadowed by the spotlight on algorithms and data. This oversight is particularly problematic in contexts like ML-as-a-service platforms, where users often lack control over the hardware used for model deployment. How does the choice of hardware impact generalization properties? This paper investigates the influence of hardware on the delicate balance between model performance and fairness. We demonstrate that hardware choices can exacerbate existing disparities, attributing these discrepancies to variations in gradient flows and loss surfaces across different demographic groups. Through both theoretical and empirical analysis, the paper not only identifies the underlying factors but also proposes an effective strategy for mitigating hardware-induced performance imbalances.

Predict-Then-Optimize by Proxy: Learning Joint Models of Prediction and Optimization

Many real-world decision processes are modeled by optimization problems whose defining parameters are unknown and must be inferred from observable data. The Predict-Then-Optimize framework uses machine learning models to predict unknown parameters of an optimization problem from features before solving. Recent works show that decision quality can be improved in this setting by solving and differentiating the optimization problem in the training loop, enabling end-to-end training with loss functions defined directly on the resulting decisions. However, this approach can be inefficient and requires handcrafted, problem-specific rules for backpropagation through the optimization step. This paper proposes an alternative method, in which optimal solutions are learned directly from the observable features by predictive models. The approach is generic, and based on an adaptation of the Learning-to-Optimize paradigm, from which a rich variety of existing techniques can be employed. Experimental evaluations show the ability of several Learning-to-Optimize methods to provide efficient, accurate, and flexible solutions to an array of challenging Predict-Then-Optimize problems.